docFacile backbone structure determination of human
download 
Report Transcription
Facile backbone structure determination of human
Articles Facile backbone structure determination of human membrane proteins by NMR spectroscopy npg © 2012 Nature America, Inc. All rights reserved. Christian Klammt1,2,6, Innokentiy Maslennikov1,2,6, Monika Bayrhuber1,3, Cédric Eichmann1,3, Navratna Vajpai1,3, Ellis Jeremy Chua Chiu1,2, Katherine Y Blain1, Luis Esquivies1, June Hyun Jung Kwon2, Bartosz Balana4, Ursula Pieper5, Andrej Sali5, Paul A Slesinger4, Witek Kwiatkowski1, Roland Riek1,3 & Senyon Choe1,2 Although nearly half of today’s major pharmaceutical drugs target human integral membrane proteins (hIMPs), only 30 hIMP structures are currently available in the Protein Data Bank, largely owing to inefficiencies in protein production. Here we describe a strategy for the rapid structure determination of hIMPs, using solution NMR spectroscopy with systematically labeled proteins produced via cell-free expression. We report new backbone structures of six hIMPs, solved in only 18 months from 15 initial targets. Application of our protocols to an additional 135 hIMPs with molecular weight <30 kDa yielded 38 hIMPs suitable for structural characterization by solution NMR spectroscopy without additional optimization. About 30% of the human protein-coding genes encode IMPs, which have critical roles in metabolism, regulation, transport and intercellular signaling. hIMPs are the targets of 50% of approved therapeutic drugs; however, difficulties with the manipulation of hIMPs have impeded the detailed functional and structural studies required to expedite drug development and discovery. These difficulties are associated with hIMP expression, purification, crystallization for X-ray structural studies, and isotopic labeling and resonance assignment for solution NMR spectroscopy studies. Notably, cellular prokaryotic expression systems generally lack compatible translocation machineries for hIMPs, and eukaryotic systems are expensive and difficult to handle. Consequently, only 30 structures of hIMPs are currently deposited in the Protein Data Bank (PDB). Recently, Escherichia coli–derived cell-free expression systems have proven effective in overcoming many limitations inherent to in vivo expression in prokaryotic hosts1. In the absence of compartmentalization in a hydrophobic milieu, IMPs produced in a cell-free expression system form precipitates that can be subsequently solubilized in mild detergents. We named this mode of expression precipitating cell-free (P-CF) expression1. Alternatively, inclusion of a detergent or a lipid can effect direct expression of solubilized IMPs2–5. We have extensively optimized P-CF expression for IMP production and had demonstrated efficient production of natively folded protein6. Other studies have also shown cell-free expression of fully functional G protein– coupled receptors and transporters7–11. Transverse relaxation optimized spectroscopy (TROSY)-based experiments have expanded the applicability of three-dimensional (3D) structure determination by solution NMR spectroscopy to large systems12, including micelle-bound membrane proteins13–17. The tremendous adaptability of cell-free expression makes it ideally suited to the isotopic labeling strategies used for these experiments. In particular, the cell-free combinatorial dual-labeling (CDL) strategy6 has greatly facilitated the usually laborious sequential assignment of IMP resonances. Furthermore, technological limitations in the acquisition of the requisite long-range distance constraints for 3D structure determination have been overcome thanks to the measurements of paramagnetic relaxation enhancement (PRE) caused by an exogenous or covalently bound paramagnetic group18–21 and the measurements of long-range nuclear Overhauser effects (NOE) for deuterated and selectively protonated proteins22 solubilized in deuterated detergents. Here we describe an application of a highly effective and fast strategy, which combines NMR spectroscopy and cell-free expression, to determine the structure of six hIMPs. RESULTS Expression screening and NMR spectroscopy We surveyed the hIMP proteome for favorable targets for solution NMR spectroscopy structural studies (Fig. 1a) and initially selected 15 moderately sized (<20 kDa), polytopic (two or more membrane crossings) hIMPs (Fig. 1b–f). We expressed all but one of these in our E. coli–derived P-CF system at high levels (>1 mg per 1 ml of reaction mixture) (Fig. 1c,d). The proteins were suitable for subsequent characterizations without 1Structural Biology Laboratory, Salk Institute for Biological Studies, La Jolla, California, USA. 2Joint Center for Biosciences at Gachon University of Medicine and Science, Incheon, Korea. 3Laboratorium für Physikalische Chemie, ETH Zürich, Zürich, Switzerland. 4Clayton Foundation Laboratories for Peptide Biology, Salk Institute for Biological Studies, La Jolla, California, USA. 5Department of Bioengineering and Therapeutic Sciences, Department of Pharmaceutical Chemistry and California Institute for Quantitative Biosciences, University of California at San Francisco, San Francisco, California, USA. 6These authors contributed equally to this work. Correspondence should be addressed to S.C. ([email protected]). Received 20 January; accepted 25 April; published online 20 May 2012; doi:10.1038/nmeth.2033 834 | VOL.9 NO.8 | AUGUST 2012 | nature methods G att2 1A D D O IG H * Solution structure 15 12 9 2 G 2 G X 3 G X 2 F 12 9 6 2 0 7 3 1 5 2 G X 4 G X 4 G e 15 6 3 3 G X 15 0 15 2 2 4 F 3 F f 1 0 2 G 15 12 9 3 P 4 F 9 5 Poor d 2 G X Fair 11 TMH: 2 NMR: F STR: Good Accelerated assignment * * * 6 3 0 0 1 Figure 1 | Workflow for screening and evaluation of 15 selected hIMPs for structural studies. (a) hIMP genes (G) are cloned into suitable Gateway-adapted cellfree vectors (Online Methods). P-CF expression is screened using a 24-well preparative high-throughput (pHTP) setup, followed by verification of expression, efficiency of detergent solubilization and suitability for NMR spectroscopy. Structure determination involves P-CF expression of uniformly or selectively labeled sample, protein solubilization, assignment using CDL and standard approaches, and collection of PRE and NOE data for structure calculation. (b,c) Western blot (b) and Coomassie stain gel (c) of indicated proteins (hIMPs with determined structures are labeled in green). TMH, numbers of predicted transmembrane helices; NMR, NMR spectral quality information labeled as good (G), fair (F) or poor (P); structure, ‘X’ marks hIMPs with determined solution structure. Asterisks indicate the position of expressed proteins. (d–f) Summary of evaluation for the 15 proteins. Expression of all 15 proteins was verified by western blot. Number of proteins classified according to cell-free expression levels (d). Number of proteins solubilized with each of the tested detergents: 70 mM sodium dodecyl sulfate (SDS), 42 mM 1-myristoyl-2-hydroxy-sn-glycero-3-[phospho-rac-(1-glycerol)] (LMPG), 100 mM n-decylphosphocholine (FC10), 100 mM n-dodecylphosphocholine (FC12), 250 mM n-decyl-β-d-maltoside (DM), mixture of 196 mM n-dodecyl-β-d-maltoside (DDM) with 41 mM cholesteryl hemisuccinate (CHS) and 100 mM lauryldimethylamine-N-oxide (LDAO) (e). Number of proteins classified according to NMR spectral quality (f). additional purification. The lipid-derived detergent 1-myristoyl2-hydroxy-sn-glycero-3-[phospho-rac-(1-glycerol)] (LMPG) was the most effective of the seven detergents screened for solubilizing the protein precipitates produced by P-CF expression (Fig. 1e). We measured [1H-15N]TROSY–heteronuclear single quantum coherence (HSQC) spectra from the solubilized, uniformly 15N-labeled hIMPs and evaluated their quality (good, fair or poor; Fig. 1c,f) according to the number of visible glycine backbone and tryptophan indole H-N resonances, the total number of cross-peaks, the chemical shift dispersion and the uniformity of line shapes. Characterization and NMR signal assignment of six hIMPs From our initial 15 hIMP preparations, nine produced good [1H-15N]TROSY-HSQC spectra and were suited for comprehensive NMR spectroscopy studies. For six of these nine hIMPs, we assigned 105 c G32 G85 G42 G81 G7 G15 G40 G25 110 TMEM14A G68 a HIGD1A 125 G37 G74 G53 G33 G34 G78 T25 T5 S11 S83 S18 120 125 T3 S50 T54 T76 Q68 S57 S86 S18 F87 E86 M75 R85 K49 I21 R22 V32 V30 N53 E13 V65 Y43 V7 A39 W88 I34 K47 V77 V71 R64 L48 L20 R22 I40 R51 S8 A38 V41 R22 Y84 F70 L45 Y81 M56 V72 D4 K55 F29 I58 L60 A89 A35 K92 A66 A24 9.8 9.7 9.6 W88 128 T64 C71 S51 S54 S28 S2 V83 tag S28 S74 S87 S16 R93 V46 L96 M67S47 I30 V35 I77 V95 S56 F61 F18 L98 F58 T13 M67 I91 M1 V12 M90 Y44 I4 L63 tag L94 Y20 K76 F34 L11 R45 A64 L92 I66 L38 F71 V55 D49 V69 F14 M78 R51 K50 A9 L82 D52 D2 L99 R73 K72 I17 V26 L3 R70 L86 R23 C37 A10 V53 K21 A39 F62 L29 F6 A31 K75 L33 A43 L57 K54 A60 L29 A84 A80 T55 E27 I62 N4 R48 V66 V40 R50 Y44 T83 C16 V81V89 K24 V41 V31 Q69 M92 L21 V73 Q94 D15 R45 C38 R26 H63 I46 R91 L22 R23 D13 K20 Y85 A43 L30 K90 M77 E14 I76 A68 L49 L61 D87 L39 M57 Y82 E19 W7 Y47 H60 E98 R5 M84 I33 R52 A93 I59 A42 A70 R6 L35 Y88 A72 V17 A75 R65 A67 7.0 6.8 K56 A3 118 W8 10.3 10.2 A96 A80 L78 W7 W8 K99 8.8 8.6 8.4 8.2 8.0 7.8 7.6 1 105 122 G21 Figure 2 | NMR spectral quality and N-H backbone assignment for six hIMPs. (a–f) [1H,15N]TROSY-HSQC spectra with assignment of NH cross-peaks for six hIMPs (Online Methods). Insets, tryptophan indole NH cross-peaks (bottom left), and high-field shifted backbone NH cross-peaks (bottom right). G5b G77 T37 TMEM141 S5 S32 S9 N58 S100 S80 A63 130 f G94 G90 G49 G45 G25 G41 120 125 T104 W62 W90 T68 T4 E94 S28 K16 S75 H17 S83 C26 V66 C86 S103 T37 K55 R54 M32 V13 M50 Q53 Q97 I52 F46 N88 L91 Q49 Q27 V74 A15 V81 V68 K85 F31 D11 F36 A12 K33 F51 W90 Y58 V9 L65 V78 L93 Y23 R8 L6 N87 M44 A25 K100 D101 R8b R80 A67 A24 H29 E22 A14 V2 E82 Y76 F92 Q106 A70 L4 N3 R102 V73 A12 L48 R107 S108 D105 L60 E84 L64 F38 L98 L20 L4 V35 A30 10.1 9.9 W62 S7 G24 G84 G70 T79 129 S34 I75 H11 10.2 10.1 W62 8.8 8.6 8.4 8.2 8.0 7.8 7.6 1 S5 S6 S56 S83 S23 S96 S106 tag N61 V30 S69 S46 V64 F108 M76 I92 T71 F13 V21 Y53 M99 I27 N109 W12 V100 L43 M78 L26 L20 V7 M107 tag D3 A67 W63 Q57 K31 A18 A40 Q54 Y29 L66 A39 A73 L91 Y17 L55 L50 A93 L10 F65 L97 V35 F15 V62 I75 L72 A95 A32 A48 A52 R110 A22 A48 tag A89 L42 L38 H112 R60 V105 W12 130 F44 G33 G14 G28 G104 G16 T40 S63 115 TMEM14C G41 G74 G51 G34 G96 S52 T82 S73 T92 129 G5a G43 G19 G88 G20 G6 tag T53 T26 S38 S103 I56 M34 V14 M51 L94 S66 V17 I86 V68 S45 L98 Q72 L31 V24 V18 V29 tag K49 V85 L28 V81 A77 M50 I70 F36 E4 V74A55 V62 R11 W97 V21 A12 A43 A69 A16 D8 L71 A15 A48 W7 A90 F89 V39 A13 A76 A64 L67 I41 K84 V27 A47 I46 L79 A57 S104 A96 A93 A30 tag A54 A33 A42 10.2 10.1 A23 T95 G47 110 V9 8.6 8.4 8.2 8.0 7.8 7.6 7.4 7.2 H (p.p.m.) e G71 G99 tag G59 S83 W97 120 130.5 tag G35 G87 G61 W7 S58 130.0 L9 129 15 N (p.p.m.) G6 115 Y8 L88 G97 G95 G2 T59 120 G36 G73 G80 110 G19 T65 HIGD1B G52 G60 S44 G79 G69 G17 G40 G75 G25 G5 FAM14B G78 G10 N (p.p.m.) G36 G44 b G65 G36 M89 105 d G19 G24 115 15 npg © 2012 Nature America, Inc. All rights reserved. Structure Determination CDL NMR (iv) Dialyzer with LMPG sequential samples hIMP precipitate ( H N C), PRE, NOE 15 h in RMX (1–4 ml) 15 13 2 15 ( N C), ( H N) (d27-LMPG) (v) data 15 ( N), (CDL), 15 (selectively N) 2 15 13 * * * * DM Detergent solubilization * * * DDM+CHS LDAO LMPG * FC12 Preparative scale NMR suitability * * * 14 FC10 (iii) 16 SDS 15 h 11 LMPG NMR samples 15 ( N) TROSYHSQC c 0.5 mg ml–1 detergent- Reaction mixture solubilization Isotopic labeling Feeding mixture (0.75–1.4 ml) Dialysis (RMX) (50–100 µl) screen (uniform, CDL) membrane Expression level, detergent solubilization 1 mg ml–1 SDS gel 14 2 mg ml–1 (24 × 70 µl) (ii) Expression verification western blot >3 mg ml–1 Screening Anti His6 (i) MW (kDa) 24-well pHTP scale Expression screening MW (kDa) 16 Number of hIMPs Cloning att1 G att2 Thr Xa Stagll His6 p23-GWN (NMR) for NMR screening hIMPencoding gene in cell-free vector AP b p23-GWT (tagged) for verification BL C a LP P1 H IG D 1 TM B EM PS 14A EN EN FA M 14 TM B EM 1 TM 41 EM 1 TM 4C EM 14 B LE PR O SC TL 1 D 5 AT P5 G 1 C N IH 4 C 14 or f1 Articles 129.0 129.5 H82 A93 8.6 8.4 8.2 8.0 7.8 7.6 7.4 7.2 H (p.p.m.) nature methods | VOL.9 NO.8 | AUGUST 2012 | 835 npg © 2012 Nature America, Inc. All rights reserved. Articles Figure 3 | Solution NMR spectroscopy structures and long-distance constraints used in structure calculations. (a) Stereo view of the TMEM14A structures calculated using separate sets of NOE- and PRE-based longrange distance constraints. The best structure calculated using NOE constraints (red) was superimposed with the best 20 structures calculated using PRE-based constraints (black). Long-range NOE constraints are shown with blue dotted lines between heavy atoms (instead of attached protons) for clarity of the figure. The structures are superimposed by backbone atoms of transmembrane helical regions. The backbone of the region Gly24–Leu98 is shown. Side chains with the long-range NOE contacts are also shown for NOE-based structure. (b) Ribbon representation of solution NMR spectroscopy structures of six hIMPs and long-distance constraints used in structure calculation. The long-range distance constraints obtained from PREs measurements are indicated by straight lines colored according to the color of the spinlabeled transmembrane helix (first is colored in green, second in blue and third in orange). Additional non-transmembrane helices and PRE constraints obtained with the spin labels located outside the transmembrane regions are colored in gray. The structures are shown in pairs with 90° vertical rotation. a resonances (Fig. 2) using the CDL strategy6 (Supplementary Table 1) and conventional sequential assignment. The percentages of resonances assigned for backbone and side-chain atoms for the six proteins are summarized in Supplementary Table 2. Static light scattering coupled with size-exclusion gel chromatography and refracting index measurements revealed that these hIMPs were monomeric in LMPG micelles (Supplementary Fig. 1). We verified that H1GD1B was also monomeric in three different detergents: LMPG, dodecylphosphocholine (FC12) and n-dodecyl beta-maltoside (DDM) (Supplementary Fig. 2). We also verified that proteins were micelleembedded and determined the regions of the target proteins embedded into the LMPG micelle by analyzing the PRE effect caused by the hydrophilic spin-labeled probe (Supplementary Fig. 3). NMR spectroscopy structure determination of hIMPs We used the chemical shift index calculated for 13Cα, 13Cβ and 13CO atoms to localize the helical regions of the proteins. We calculated backbone structures of the six hIMPs based on long-range distance constraints obtained from PRE measurements of the spinlabeled proteins using single cysteine mutants (Supplementary Figs. 4 and 5 and Supplementary Table 3). To validate the PREbased structure calculations, we calculated the spatial structure of TMEM14A using PRE and independently using long-range distance constraints obtained from NOEs. The structures were similar, with average r.m.s. deviation between the backbone atoms of the transmembrane helical regions in the PRE- and NOE-based structures equal to 3.05 Å (Fig. 3a). The NOE-based structure of TMEM14A comprised a more tightly packed helical bundle than the PRE-based structure, whereas the orientation and topology of the transmembrane helices were very similar. The reason for the less tight packing of the bundle in the PRE-based structure 836 | VOL.9 NO.8 | AUGUST 2012 | nature methods b HIGD1A HIGD1B TMEM14A TMEM141 TMEM14C FAM14B lies in the impossibility of calculating short (<12 Å) PRE constraints using nitroxide paramagnetic spin label and in the lower precision of this type of constraints as compared to NOE-based constraints (Online Methods). To determine whether the helical orientation can be unambiguously determined from the PRE data, we calculated the PRE constraints error for generic structures with transmembrane helices rotated around the helical axes. The structures with the low cumulative error function (smaller than 1 Å2) calculated for PRE constraints were located within a 20° rotation range from the starting structure (angles 0, 0) (Supplementary Fig. 6). PRE distances quality factor did not exceed 13.6% among the calculated hIMP structures (Table 1), which shows good agreement between PRE-derived distances used in structure calculation and distances back-calculated from the structures. NMR spectroscopy experimental data, including structural and refinement statistics for the calculated structures, are summarized in Table 1. The long-range distance constraints used for structure calculation are illustrated in Figure 3b. NMR spectroscopy structures of six hIMPs All six hIMP structures were helical bundles with helix lengths and exposed hydrophobic faces consistent with the bilayer-embedded localization of these proteins (Fig. 4). In agreement with the numbers of predicted transmembrane helices (Fig. 1c), HIGD1A, HIGD1B and TMEM141 had two transmembrane crossings, whereas TMEM14A had three transmembrane crossings. compared to the prediction of three and four transmembrane helices, FAM14B and TMEM14C had two and three transmembrane crossings, respectively (Fig. 4 and Supplementary Fig. 7). The first transmembrane helix of both FAM14B and TMEM141 was severely kinked (by about 50°; Articles Table 1 | Summary of NMR spectroscopy data and statistics for the calculated sets of 20 lowest-energy structures npg © 2012 Nature America, Inc. All rights reserved. TMEM14A Protein size (amino acids) Transmembrane segments (predicteda) Global rotational correlation timeb (ns) Cysteine mutants NMR constraints NOE (long-range NOEd) PRE: upper/lower Number of PRE per restrained residue: upper/total restrained distances Hydrogen bonds Dihedral angle: Phi/Psi Structure statisticse Violations (mean ± s.d.) Distance, sum (Å) Distance, average (Å) Maximum distance violation (Å) Dihedral angle, sum (°) Maximum dihedral angle violation (°) PRE distances Q-factor (%) Average backbone r.m.s. deviation (Å) Ramachandran plot statisticse,f Most favored regions (%) Additionally/generously allowed (%) Disallowed (%) Deviations from idealized geometryg Bond length (Å) Bond angles (°) Equivalent resolutionf,g (Å) HIGD1A HIGD1B NOE FAM14B TMEM141 TMEM14C 93 2 (2) 99 2 (2) 99 3 (3) PRE 104 2 (3) 108 2 (2) 112 3 (4) 25.1 28.8 25.5 21.6 20.7 25.5 6 6 7 8 5 9 40c 156/186 3.63/5.23 47c 224/240 4.07/6.34 651 (18d) 0 0 0 334/467 4.64/7.38 49c 195/389 3.98/8.39 153 162/235 1.80/2.61 55c 283/479 4.56/8.76 41 43/43 44 53/55 44 72/72 44 72/72 51 45/49 60 86/90 58 62/61 2.80 ± 0.10 0.19 ± 0.03 0.24 ± 0.01 1.35 ± 0.11 0.17 ± 0.04 0.25 ± 0.03 2.53 ± 0.39 0.15 ± 0.06 0.23 ± 0.03 2.42 ± 0.18 0.13 ± 0.06 0.21 ± 0.03 3.11 ± 0.16 0.26 ± 0.14 0.51 ± 0.02 1.15 ± 0.07 0.09 ± 0.04 0.14 ± 0.01 3.29 ± 0.36 0.21 ± 0.06 0.31 ± 0.04 1.36 ± 0.55 0.56 ± 0.13 1.02 ± 0.67 0.51 ± 0.30 3.78 ± 1.24 0.90 ± 0.29 2.11 ± 0.60 0.84 ± 0.32 4.87 ± 0.72 0.75 ± 0.19 3.73 ± 1.37 0.48 ± 0.10 4.78 ± 1.39 0.73 ± 0.21 12.3 1.52 ± 0.59 13.4 0.73 ± 0.19 NA 1.72 ± 0.65 12.1 1.82 ± 0.48 13.6 1.11 ± 0.42 NA 3.52 ± 1.51 13.1 1.42 ± 0.37 71.6 19.6/6.2 67.3 23.1/7.1 73.0 17.6/7.1 70.6 22.4/4.9 70.7 21.9/4.8 80.9 15.2/2.8 71.9 17.8/7.2 2.5 2.5 2.3 2.1 2.7 1.1 3.1 0.013 1.6 3.3 0.012 1.6 3.3 0.013 1.5 2.8 0.013 1.5 3.1 0.010 1.6 3.2 0.012 1.6 3.0 0.013 1.5 3.1 aNumber of predicted transmembrane helices is given in parentheses. bCalculated for total mass of protein-detergent complex using the Stokes-Einstein equation. cInter-residue sequential (|i – j| = 1) constraints only. dNumber of long-range (|i – j| > 4) NOE constraints is given in parentheses. eCalculated for transmembrane bundles. fCalculated by Procheck program31. gBased on Ramachandran plot quality assessment. NA, not applicable. Supplementary Table 4). For HIGD1A, HIGD1B, TMEM141 and TMEM14A, the first transmembrane helix was preceded by an amphiphilic N-terminal helix; this helix is presumably located at the micelle-water interface, but we could not define its exact orientation relative to the transmembrane helical bundles. The connecting loops between the transmembrane helices of FAM14B and second and third transmembrane helices of TMEM14C contain an amphiphilic helix, which lies roughly perpendicular to the preceding transmembrane helix. As a moderate PRE effect caused by the soluble paramagnetic agent was detected for backbone amides of the amphiphilic helices (Supplementary Fig. 3), we can assume that the helices are located close to the surface of the LMPG micelle. TMEM141 has remarkably elongated transmembrane helices of 34 and 33 amino acids. The N-terminal part of the first transmembrane helix and C-terminal part of the second one protruded from the micelle as confirmed by a moderate PRE effect from the soluble paramagnetic agent (Supplementary Fig. 3d). The three-helical bundles of TMEM14A and TMEM14C were tightly packed with interhelical distances less than 8 Å, whereas the two-helical bundles of HIGD1A, HIGD1B, FAM14B and TMEM141 were more loosely packed with interhelical distances exceeding 8 Å and relatively few interhelical van der Waals contacts localized close to the ends of the helices. The parameters describing packing of the helices (pair-wise angles between the transmembrane helices and bending angles of the transmembrane helices) are listed in Supplementary Table 4. The backbone structures of HIGD1A, HIGD1B, FAM14B, TMEM141, TMEM14A and TMEM14C reported here account for 20% of hIMP entries currently in PDB. The structures of these proteins could immediately serve as a model for a substantial portion of the membrane proteome. A search of the UniProtKB database of all known protein sequences identified 609 unique protein sequences with sequence identity greater than 30% to at least one of the six hIMPs. Based on the determined structures, we calculated structural models for these 609 unique protein sequences. nature methods | VOL.9 NO.8 | AUGUST 2012 | 837 Articles HIGD1A HIGD1B FAM14B TMEM141 TMEM14C npg © 2012 Nature America, Inc. All rights reserved. TMEM14A Figure 4 | Solution NMR spectroscopy structures of six hIMPs. Structures were calculated by Cyana using distance information obtained from NOE and PRE measurements. The first transmembrane helix is in green, second in blue and third in orange. Additional non-transmembrane helices are in gray. Shown are ribbon structures (left), superposition of the best 20 backbone structures (middle) and their 90° vertical rotation (right). Additional lower accuracy structural information is provided for additional 380 protein sequences at sequence identities below 30% (Supplementary Table 5). hIMP-specific antibody generation by a P-CF product We demonstrated that the P-CF–expressed hIMP elicit the production of highly specific polyclonal antibodies. A rabbit polyclonal antibody (Eton Bioscience) was generated to P-CF–expressed and detergent-solubilized HIGD1A. The anti-HIGD1A antibody preparation recognized an overexpressed HIGD1A-GFP fusion in HEK293T cells and endogenously expressed HIGD1A in both 838 | VOL.9 NO.8 | AUGUST 2012 | nature methods HEK293T cells and hippocampal neurons but did not recognize the homologous (43% sequence identical) protein HIGD1B (Supplementary Fig. 8). Screening a broad range of hIMPs The success of our preliminary studies spurred a more extensive coverage of the hIMP proteome. From a library of 3,270 hIMPs23, we selected an additional 135 targets in the 10–30 kDa range for P-CF expression, solubilization screening and preliminary NMR spectroscopy analysis (Supplementary Fig. 9). Overall, 111 (74%) of 150 targets expressed at considerably high levels (>1 mg ml−1 of cell-free reaction mixture; Supplementary Fig. 10 and Supplementary Table 6), and we used LMPG to solubilize all of them. From an analysis of [1H-15N]TROSY-HSQC spectra, we found that 38 of 100 evaluated by NMR spectroscopy targets, including the six hIMPs with solved structure, were adequate for structural studies without additional optimization (Supplementary Figs. 11 and 12). DISCUSSION The described structural studies of hIMPs demonstrate the technological synergy between cell-free expression and CDL-aided NMR spectroscopy analysis. It has been shown previously that PRE-based constraints can be used for structure determination of β-barrel membrane proteins19 and small single-crossing helical membrane proteins18. They also have been used as additional long-range distance constraints for structure determination of multihelical membrane proteins14,24,25. Our results demonstrate that PRE-derived distances can be used as a single source of long-range constraints for structure determination of multihelical membrane proteins. A careful design of the mutants for spin labeling and a meticulous analysis of PRE data, however, are necessary because changes in protein conformation induced by mutation, high mobility of the spin-labeled residue and sparse net of long-range PRE data may affect accuracy of the structures. The biological functions of the six hIMPs we characterized here have not yet been fully defined, and we do not know whether their structures are in the functionally relevant state. Nevertheless, knowledge of the backbone scaffolds of the proteins may provide structural insights for, for example, site-specific mutagenesis, which would help understand the fundamental functional roles of these proteins. HIGD1A and HIGD1B are likely associated with response to hypoxia26. HIGD1A has been found to be upregulated in hypoxia27; HIGD1B in prolactinomas is speculated to be associated with increased tumor hypoxia tolerance, angiogenesis and drug resistance28. FAM14B, a member of the FAM14 family encoded by interferon-stimulated gene 12c is localized in the mitochondria and may influence cellular sensitization to apoptotic stimuli via mitochondrial membrane destabilization29. Distinct from TMEM141 and TMEM14A, TMEM14C belongs to an uncharacterized protein family UPF0136_TM that is presumably involved in heme biosynthesis30. Consistently with our earlier speculation6, we predict that IMPs with tightly packed helices (TMEM14A and TMEM14C) may have a structural or transport role in the membrane, whereas IMPs with loosely packed helices (HIGD1A, HIGD1B and TMEM141) could be involved in signal transduction across the membrane. We believe that efficient production of hIMPs by cell-free expression in combination with robust structural analysis by NMR Articles spectroscopy has broader applications. In addition to facilitating 3D structure determination of membrane proteins by solution NMR spectroscopy, it will benefit functional and biochemical characterization of hIMPs, including individual antibody generation against hIMPs for proteomic and cell biological studies. Engineered hIMPs could also constitute a building block in biomaterial and nanoscience research. Methods Methods and any associated references are available in the online version of the paper. Accession codes. PDB: HIGD1A, 2LOM; HIGD1B, 2LON; FAM14B, 2LOQ; TMEM141, 2LOR; TMEM14A, 2LOO and 2LOP; and TMEM14C, 2LOS. npg © 2012 Nature America, Inc. All rights reserved. Note: Supplementary information is available in the online version of the paper. Acknowledgments We thank G. Louie for comments in preparation of the manuscript, A.S. Arseniev for suggestions on the spin-labeling procedure and S. Maslennikov for writing the atomDistancer program. C.K. thanks the Pioneer Foundation for a Pioneer Fund Postdoctoral Scholar Award. This work has been partly supported by US National Institutes of Health (S.C.: GM098630, GM095623; A.S. and U.P.: GM094662, GM094625 FDP, and GM54762), Incheon Free Economic Zone and the World Class University Program (Korea). AUTHOR CONTRIBUTIONS C.K., I.M., W.K., R.R. and S.C. designed experiments, C.K., E.J.C.C., L.E. and J.H.J.K. cloned hIMP targets, performed cell-free expression, evaluated protein expression levels and detergent solubilization; C.K. and E.J.C.C. created single cysteine mutants for PRE experiments, prepared isotopically labeled NMR spectroscopy samples and samples for PRE measurements. C.K. and I.M. recorded NMR spectra and evaluated NMR spectral quality of tested hIMPs; C.K., I.M., M.B., C.E., N.V., E.J.C.C. and K.B. collected and assigned NMR spectra and analyzed data; I.M., M.B. and C.E. calculated the structures. C.K., E.J.C.C., B.B. and P.A.S. analyzed HIGD1A antibody specificity by western blot and by immunostaining; U.P. and A.S. calculated modeling leverage based on hIMP structures; C.K., I.M., W.K. and S.C. wrote the manuscript. All authors discussed the results and commented on the manuscript. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. Published online at http://www.nature.com/doifinder/10.1038/nmeth.2033. Reprints and permissions information is available online at http://www.nature. com/reprints/index.html. 1. Klammt, C. et al. High level cell-free expression and specific labeling of integral membrane proteins. Eur. J. Biochem. 271, 568–580 (2004). 2. Wuu, J.J. & Swartz, J.R. High yield cell-free production of integral membrane proteins without refolding or detergents. Biochim. Biophys. Acta 1778, 1237–1250 (2008). 3. Katzen, F. et al. Insertion of membrane proteins into discoidal membranes using a cell-free protein expression approach. J. Proteome Res. 7, 3535–3542 (2008). 4. Kalmbach, R. et al. Functional cell-free synthesis of a seven helix membrane protein: in situ insertion of bacteriorhodopsin into liposomes. J. Mol. Biol. 371, 639–648 (2007). 5. Klammt, C. et al. Evaluation of detergents for the soluble expression of alphahelical and beta-barrel-type integral membrane proteins by a preparative scale individual cell-free expression system. FEBS J. 272, 6024–6038 (2005). 6. Maslennikov, I. et al. Membrane domain structures of three classes of histidine kinase receptors by cell-free expression and rapid NMR analysis. Proc. Natl. Acad. Sci. USA 107, 10902–10907 (2010). 7. Klammt, C. et al. Polymer-based cell-free expression of ligand-binding family B G-protein coupled receptors without detergents. Protein Sci. 20, 1030–1041 (2011). 8. Junge, F. et al. Modulation of G-protein coupled receptor sample quality by modified cell-free expression protocols: a case study of the human endothelin A receptor. J. Struct. Biol. 172, 94–106 (2010). 9. Keller, T. et al. Cell free expression and functional reconstitution of eukaryotic drug transporters. Biochemistry 47, 4552–4564 (2008). 10. Klammt, C. et al. Functional analysis of cell-free-produced human endothelin B receptor reveals transmembrane segment 1 as an essential area for ET-1 binding and homodimer formation. FEBS J. 274, 3257–3269 (2007). 11. Ishihara, G. et al. Expression of G protein coupled receptors in a cell-free translational system using detergents and thioredoxin-fusion vectors. Protein Expr. Purif. 41, 27–37 (2005). 12. Pervushin, K., Riek, R., Wider, G. & Wuthrich, K. Attenuated T2 relaxation by mutual cancellation of dipole-dipole coupling and chemical shift anisotropy indicates an avenue to NMR structures of very large biological macromolecules in solution. Proc. Natl. Acad. Sci. USA 94, 12366–12371 (1997). 13. Gautier, A. et al. Structure determination of the seven-helix transmembrane receptor sensory rhodopsin II by solution NMR spectroscopy. Nat. Struct. Mol. Biol. 17, 768–774 (2010). 14. Van Horn, W.D. et al. Solution nuclear magnetic resonance structure of membrane-integral diacylglycerol kinase. Science 324, 1726–1729 (2009). 15. Wang, J., Pielak, R.M., McClintock, M.A. & Chou, J.J. Solution structure and functional analysis of the influenza B proton channel. Nat. Struct. Mol. Biol. 16, 1267–1271 (2009). 16. Hiller, S. et al. Solution structure of the integral human membrane protein VDAC-1 in detergent micelles. Science 321, 1206–1210 (2008). 17. Bayrhuber, M. et al. Structure of the human voltage-dependent anion channel. Proc. Natl. Acad. Sci. USA 105, 15370–15375 (2008). 18. Page, R.C. et al. Backbone structure of a small helical integral membrane protein: A unique structural characterization. Protein Sci. 18, 134–146 (2009). 19. Liang, B., Bushweller, J.H. & Tamm, L.K. Site-directed parallel spinlabeling and paramagnetic relaxation enhancement in structure determination of membrane proteins by solution NMR spectroscopy. J. Am. Chem. Soc. 128, 4389–4397 (2006). 20. Iwahara, J., Schwieters, C.D. & Clore, G.M. Ensemble approach for NMR structure refinement against (1)H paramagnetic relaxation enhancement data arising from a flexible paramagnetic group attached to a macromolecule. J. Am. Chem. Soc. 126, 5879–5896 (2004). 21. Battiste, J.L. & Wagner, G. Utilization of site-directed spin labeling and high-resolution heteronuclear nuclear magnetic resonance for global fold determination of large proteins with limited nuclear overhauser effect data. Biochemistry 39, 5355–5365 (2000). 22. Kay, L.E. Solution NMR spectroscopy of supra-molecular systems, why bother? A methyl-TROSY view. J. Magn. Reson. 210, 159–170 (2011). 23. Rual, J.F. et al. Human ORFeome version 1.1: a platform for reverse proteomics. Genome Res. 14, 2128–2135 (2004). 24. Sobhanifar, S. et al. Structural investigation of the C-terminal catalytic fragment of presenilin 1. Proc. Natl. Acad. Sci. USA 107, 9644–9649 (2010). 25. Zhou, Y. et al. NMR solution structure of the integral membrane enzyme DsbB: functional insights into DsbB-catalyzed disulfide bond formation. Mol. Cell 31, 896–908 (2008). 26. Bedo, G. et al. Characterization of hypoxia induced gene 1: expression during rat central nervous system maturation and evidence of antisense RNA expression. Int. J. Dev. Biol. 49, 431–436 (2005). 27. Kasper, L.H. & Brindle, P.K. Mammalian gene expression program resiliency: the roles of multiple coactivator mechanisms in hypoxiaresponsive transcription. Cell Cycle 5, 142–146 (2006). 28. Jiang, Z., Gui, S. & Zhang, Y. Analysis of differential gene expression by fiber-optic BeadArray and pathway in prolactinomas. Endocrine 38, 360–368 (2010). 29. Cheriyath, V., Leaman, D.W. & Borden, E.C. Emerging roles of FAM14 family members (G1P3/ISG 6–16 and ISG12/IFI27) in innate immunity and cancer. J. Interferon Cytokine Res. 31, 173–181 (2011). 30. Nilsson, R. et al. Discovery of genes essential for heme biosynthesis through large-scale gene expression analysis. Cell Metab. 10, 119–130 (2009). 31. Laskowski, R.A., MacArthur, M.W., Moss, D.S. & Thornton, J.M. PROCHECK: A program to check the stereochemical quality of protein structures. J. Appl. Cryst. 26, 283–291 (1993). nature methods | VOL.9 NO.8 | AUGUST 2012 | 839 © 2012 Nature America, Inc. All rights reserved. npg ONLINE METHODS Expression vector design. To enable cloning of hIMPs from a library of 3,270 hIMPs in Gateway entry vectors23, the pIVEX2.3d vector (Roche Applied Science) was Gateway-adapted and optimized. We designed two vectors, one containing several tags that allow detection and purification and one without tag, which was used for NMR spectroscopy sample preparation. pIVEX 2.3d was supplemented with a 5′ attR1 site (5′-ACAA GTTTGTACAAAAAAGCAGGCTTA-3′) and a 3′ attR2 site (5′-GACCCAGCTTTCTTGTACAAAGTGGTT-3′), a thrombin cleavage site (5′-GCTGCCACGCGGCACCAG-3′), a factor Xa cleavage site (5′-ATCGAGGGCCGT-3′) and a StrepII tag (5′-TGGAGCCACCCGCAGTTCGAAAAA-3′) using suitable oligonucleotide primers with suitable restriction sites and standard polymerase chain reaction techniques with Vent DNA polymerase (New England Biolab (NEB)) following standard protocols for Gateway destination vector creation (Invitrogen). The resulting pIVEX2.3d-Gateway-tag vector (p23-GWT) encodes a protein with an N-terminal Gateway sequence (MTSLYKKVG) and a C-terminal tag (Y(or C)PTFLYKVVLVPRGSHMIEGRWSHPQ FEKYRAPGGGSHHHHHH) (Fig. 1a). For Gateway cloning of nontagged hIMPs for NMR spectral quality evaluation, a second vector pIVEX2.3d-Gateway-NMR (p23-GWN) was derived from p23-GWT by introducing a stop codon (TAA) after the 5′ att site, resulting in translation of a short 9-amino-acid C-terminal attderived sequence (Y(or C)PTFLYKVV). Cloning procedures. One hundred fifty hIMP targets in Gateway entry vectors (Supplementary Table 6) were cloned into p23GWT and p23-GWN destination vectors using LR Clonase (Invitrogen) according to the manufacturer’s protocol with optimizations. In particular, 1 µl entry clone (150 ng/µl), 1 µl destination vector (150 ng/µl), 2 µl 5× LR Clonase reaction buffer, 4 µl TE buffer (pH 8.0) and 2 µl Gateway LR Clonase II enzyme mix (Invitrogen) were mixed and incubated for 60 min at 25 °C. Subsequently, 1 µl proteinase K solution was added and LR reaction mixture was incubated for 10 min at 37 °C. One microliter of LR reaction was transformed into 10 µl of DH5α chemically competent cells (Invitrogen) and plated on LB plates containing 100 µg/ml ampicillin. Single colonies were picked and grown overnight in 5 ml TB medium with 0.1 mg/ml ampicillin and purified using a Miniprep kit (Qiagen). The manufacturer’s protocol was optimized to enhance plasmid yield and purity. In particular, cells from 5-ml overnight cultures in TB medium were resuspended in 250 µl buffer P1, and 350 µl buffer P2 was added and mixed by gentle inversion. After 5 min, 450 µl of buffer N3 was added, mixed and centrifuged for 10 min at 20,000g. The supernatant was transferred to the QIAprep spin column connected to a vacuum manifold (Qiagen). The column was washed with 500 µl buffer PB and 750 µl buffer PE and subsequently centrifuged for 1 min at 20,000g to remove residual ethanol. Plasmid DNA was eluted with 75 µl buffer EB. Plasmids were checked by DNA sequencing and used for cell-free expression. Expression constructs for NMR spectroscopy structural studies of HIGD1A, HIGD1B, TMEM14A, FAM14B, TMEM141 and TMEM14C were obtained by site-directed mutagenesis, introducing a stop codon (TAA) after the hIMP-encoding genes in corresponding p23-GWN vectors. Cysteine residues in HIGD1A, HIGD1B, TMEM14A, FAM14B, TMEM141 and TMEM14C as nature methods well as serine residues in HIGD1B, TMEM14A and TMEM141 for different cysteine constructs, were introduced by site-directed mutagenesis. In particular, primers were designed as described elsewhere32 and quick-change reactions were carried out using 1 µl HotStar polymerase (Qiagen), 1× HotStar buffer, 2% DMSO, 0.2 µM primers and 3–5 µg/ml template DNA in 50 µl of reaction volume. PCR was set up in a thermocyler (Techne) at 95 °C for 0.5 min and cycled 18 times at 95 °C for 0.5 min, 55 °C for 100 s, 68 °C for 10 min with the final extension time of 30 min at 68 °C. Parental DNA was digested with DpnI (NEB) by adding 1 µl enzyme, incubated for 3 h at 37 °C and subsequently purified by a Nucleotide purification kit (Qiagen) with elution in 30 µl H2O. Seven microliters of DNA was used to transform 25 µl of DH5α chemically competent cells (Invitrogen). For transient expression of HIGD1A-GFP and HIGD1B-GFP in HEK293T cells, the pTT5 expression vector33 was Gatewayadapted with suitable oligonucleotide primers as described above, resulting in the pTT5-Gateway destination vector encoding an N-terminal Human IgG κ signal peptide (METDTLLLWVLLLW VPGSTGAGS) followed by a His9 tag and C-terminal translated GFP. Sequences encoding HIGD1A and HIGD1B in Gatewayentry vectors were cloned into pTT5-Gateway as described above. Plasmids were checked by DNA sequencing and amplified by HighSpeed Plasmid Maxi Kit (Qiagen). Cell-free expression. We established, optimized and fine-tuned for expression of IMPs a preparative high-throughput E. coli– based cell-free expression system. Chemicals for cell-free expression were purchased from Sigma-Aldrich. Stable isotope–labeled amino acids and amino acid mixtures were purchased from Cambridge Isotope Laboratories unless otherwise stated. hIMPs were produced in an individual continuous exchange cell-free system according to previously described protocols with additional optimization. In short, cell-free extracts were prepared from the E. coli strain A19 as described previsously1,34, T7 RNA polymerase was expressed using the pT7-911Q plasmid35 and purified as described previously36. Analytical-scale reactions were performed in 20 kDa molecular weight cutoff (MWCO) Mini SlideA-Lyzers (Thermo Scientific) using 70 µl of reaction mixture and 1 ml of feeding mixture. Mini Slide-A-Lyzers were placed in a custom made 24-well plastic block holding the feeding mixture and incubated in a shaker (New Brunswick Scientific) (Fig. 1a) for ~15 h at 30 °C at 160 r.p.m. Preparative scale cell-free reactions were performed in 20-kDa MWCO Slide-A-Lyzers (Thermo Scientific) using 2–4 ml of reaction mixture set with the 1:17 volume ratio between reaction and feeding mixture. Slide-A-Lyzers were placed in a suitable plastic box holding the feeding mixture and incubated in a shaker (New Brunswick Scientific) for ~15 h at 30 °C at 140 r.p.m. The conditions for the cell-free reaction were as follows: reaction mixture and feeding mixture contained 230 mM potassium acetate, 13 mM magnesium acetate, 100 mM Hepes-KOH pH 8.0, 3.5 mM Tris-acetate pH 8.2, 0.2 mM folinic acid, 0.05% sodium azide, 2% polyethyleneglycol 8000, 2 mM Tris(2-carboxyethyl) phosphine hydrochloride (TCEP) (Thermo Scientific), 1.2 mM ATP, 0.8 mM each of CTP, UTP and GTP, 20 mM acetyl phosphate (Fluka), 20 mM phosphoenol pyruvate (AppliChem), 1 tablet per 50 ml complete protease inhibitor (Roche Applied Science), 1 mM each amino acid, 40 µg/ml pyruvate kinase (Roche Applied Science), 500 µg/ml E. coli tRNA doi:10.1038/nmeth.2033 npg © 2012 Nature America, Inc. All rights reserved. mix (Roche Applied Science), 0.3 unit/µl RNase inhibitor (SUPERase-In, Ambion), 0.5 unit/µl T7 RNA polymerase, 40% S30 extract and 10 µg/ml of pIVEX2.3d–derived plasmid DNA. For cellfree uniform 15N labeling, reaction mixture and feeding mixture were supplemented with 0.5 mM of [15N]algal amino acid mixture and 0.5 mM of 15N-labeled amino acids Asn, Cys, Gln and Trp. For cell-free uniformly 15N-13C, 2H-15N, 2H-15N-13C labeling, reaction mixture and feeding mixture were supplemented with 0.5 mM of correspondingly labeled amino acid mixtures. For combinatorial labeling of HIGD1A, HIGD1B and FAM14B combinations of 15N-labeled Ala, Cys, Asp, Glu, Phe, Gly, Ile, Lys, Leu, Met, Asn, Gln, Arg, Ser, Thr, Val, Trp and Tyr or 1-13C-labeled Ala, Cys, Asp, Glu, Phe, Gly, Ile, Lys, Leu, Met, Pro, Gln, Ser, Val, Trp and Tyr, and nonlabeled amino acids were used according to the schemes given in Supplementary Table 1. Uniform 2H-15N and 2H-15N-13C labeling was efficiently done in H O. 2 Protein characterization. The Invitrogen gel electrophoresis system was used for all SDS-gel analyses following the manufacturer’s protocol, using 12% NuPAGE Bis-Tris gels in 2-(N-morpholino) ethanesulfonic acid (MES) buffer stained with Coomassie blue or InstantBlue (Expedeon Protein Solutions). The expression yield was quantified by estimating expression based on Coomassie-stained protein band intensities for all 150 hIMPs. These intensities were compared to Coomassie-stained standard bands of a known protein concentration. For western blot analysis of cell-free expressed hIMPs, the gels were blotted on a 0.45 µm Immobilon-P poly(vinylidene difluoride) membrane (Millipore) using Invitrogens Xcell IITM Blot Module for 1 h at 35 V. The membrane was then blocked for 1 h in blocking-buffer (1× PBS, 7% milk powder and 0.1% (w/v) Tween-20) and subsequently incubated for 1 h with horseradish peroxidase (HRP)-conjugated His6-tag antibody (ab1187, Abcam) using 1: 2,000 dilution in washing buffer (1× PBS and 0.1% Tween-20). After extensive washing in washing buffer, the blots were analyzed by chemiluminescence (ECL western blot substrate, Thermo Scientific) on X-ray film (CL-XPosure, Thermo Scientific) using exposure times of 10–60 s. For western blot analysis of HIGD1A and HIGD1A-GFP, we used polyclonal anti-HIGD1A antibody raised in rabbit (Eton Biosciences) from P-CF–expressed and LMPG-solubilized HIGD1A, or rabbit anti-GFP (full-length) antibody (sc-8334, Santa Cruz Biotechnology). The gels were blotted and blocked as described above with subsequent 1 h incubation with anti-HIGD1A-IgG using 1:2,000 dilution or anti-GFP-IgG using 1:1,000 dilution in washing buffer containing 7% milk powder. After incubation with primary antibody, the membrane was washed 5 times with 100 ml washing buffer for 5 min each time, and subsequently incubated for 1 h with secondary bovine anti-rabbit IgG-HRP (sc-2370, Santa Cruz Biotechnology) using 1:3,000 dilution in washing buffer supplemented with 7% milk powder. After five 5-min washes with 100 ml washing buffer, the blots were analyzed by chemiluminescence on X-ray film using exposure of 0.5–5 min. All cell-free expressed hIMPs were characterized by SDS-PAGE (Supplementary Fig. 10). HIGD1A, HIGD1B, TMEM14A, FAM14B, TMEM141 and TMEM14C were analyzed by light scattering coupled with size-exclusion chromatography and refracting index measurements (SECUV/LS/RI) (Supplementary Fig. 1). SEC-UV/LS/RI analysis of doi:10.1038/nmeth.2033 hIMP-LMPG complexes was performed by measuring the relative refractive index signal (Optilab rEX, Wyatt Technology), static light scattering signals from three angles (45°, 90° and 135°) (miniDAWN TREOS, Wyatt Technology), and UV-light extinction at 280 nm (Waters 996 Photoiode Array Detector, Millipore) during size-exclusion chromatography (HPLC, Waters 626 Pump, 600S Controller, Millipore) with polymer column (Shodexfi Protein KW-802.5). hIMPs were analyzed by injecting 100 µl of 200 µM hIMP solubilized in LMPG into high-performance liquid chromatography (HPLC) buffer (20 mM MES-Bis-Tris pH 6.0 and 150 mM NaCl) supplemented with 0.01% LMPG at 0.8 ml/min. The fractions, containing target proteins, were concentrated in 5 kDa MWCO Vivaspin2 concentrators (Sartorius Stedim Biotech) to 20–50 µl and reloaded on the column. The oligomeric state of HIGD1B in FC12, DDM and DM micelles was analyzed by SEC-UV/LS/RI. HIGD1B was solubilized in 100 µl of a buffer (20 mM MES-Bis-Tris pH 6.0 and 150 mM NaCl) containing selected detergent (20 mM, 30 mM and 75 mM for FC12, DDM and DM, respectively). The final protein concentration was 150–200 µM. The samples were injected into HPLC buffer supplemented with 1.6 mM FC12. The data were collected and analyzed using the Astra V 5.3.2.12 Software (Wyatt Technology Corp.). The average molar weights of the protein-detergent complex, the protein and the detergent fraction in the complex (Supplementary Fig. 1 and 2) were calculated by the Protein Conjugate module of the Astra program. The oligomeric state of HIGD1B in DDM and DM could not be derived by the Astra V Software from SEC-UV/LS/RI data due to the overlap of protein-detergent micelles with empty detergent micelles as shown for HIGD1B in the presence of DDM (Supplementary Fig. 2c). Nevertheless the elution volumes of the HIGD1B protein peak in the four detergents (Supplementary Fig. 2) are nearly identical, which suggests that HIGD1B in the presence of DDM and DM is most likely monomeric. Detergent solubilization. All 150 P-CF-expressed hIMPs were analyzed for detergent solubilization in seven different detergents. Detergent solubilization was tested in 70 mM sodium dodecyl sulfate (SDS), 42 mM 1-myristoyl-2-hydroxy-sn-glycero-3-[phosphorac-(1-glycerol)] (LMPG), 100 mM n-decylphosphocholine (FC10), 100 mM n-dodecylphosphocholine (FC12), 250 mM n-decyl-β-d-maltoside (DM), mixture of 196 mM n-dodecyl-βd-maltoside (DDM) with 41 mM cholesteryl hemisuccinate (CHS) and 100 mM lauryldimethylamine-N-oxide (LDAO). Seven samples of 7 µl of P-CF precipitate resuspended in buffer (20 mM Tris pH 7.4 and 150 mM NaCl) were centrifuged for 10 min at 20,000g. The supernatant was removed, and 7 µl of SDS, LMPG, FC10, FC12, DM, DDM/CHS and LDAO were added to the respective precipitate samples. The precipitate was resuspended by pipetting 10–20 times and incubated for 1 h at 37 °C. The residual precipitate was pelleted by centrifugation for 15 min at 20,000g and 1–4 µl of the supernatant, depending on hIMP expression, was loaded with 5 µl of 2× SDS sample buffer on a 12% NuPAGE Bis-Tris gel (Invitrogen) and run in MES buffer for 52 min at 200 V. The gel was stained with InstantBlue and analyzed for expression based on band intensity. NMR spectroscopy sample preparation. All hIMPs were expressed as precipitate (P-CF) in the absence of detergents1. The only gene expressing in the cell-free system is the target nature methods © 2012 Nature America, Inc. All rights reserved. npg protein gene; therefore, the only labeled protein is the target protein. This, in combination with the fact that the SEC-UV/LS/RI data showed our target proteins to be homogeneous protein-detergent complexes, allows us to conclude that co-precipitated endogenous cell-free extract proteins will not influence NMR structural studies. Therefore, additional purification of the proteins is not crucial. Precipitated recombinant proteins were removed from the reaction mixture by centrifugation at 20,000g for 15 min and washed in two steps. First, to remove co-precipitated RNA, precipitates were suspended in 50% volume equal to the reaction mixture volume in 20 mM MES-Bis-Tris buffer pH 6.0, 0.01 mg/ml RNase A and shaken at 900 r.p.m. and 37 °C for 30 min. After incubation, precipitates were collected by centrifugation at 20,000g for 10 min and suspended in 100% volume equal to the reaction mixture volume in NMR buffer (20 mM MES-Bis-Tris pH 6.0). NMR spectroscopy samples were prepared from washed precipitate of 1–4 ml reaction mixture by solubilization in 300 µl 3% (wt/vol) LMPG in NMR buffer for all tested hIMPs except HIGD1A, HIGD1B and TMEM141, which were solubilized in 2% LMPG (wt/vol) in NMR buffer. The suspension was sonicated in a water bath sonicator (Bransonic) for 1 min and subsequently incubated for 15 min with shaking at 900 r.p.m. and 37 °C, followed by centrifugation at 20,000g for 10 min. NMR spectroscopy samples were pH-adjusted and supplemented with 5% D2O and 0.5 mM 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS). For 13C and 15N NOESY NMR spectroscopy experiments requiring deuterated detergent, hIMP NMR spectroscopy samples were prepared by solubilization in 2% d27-LMPG (wt/vol) (FBReagents). D27-LMPG was used to minimize the spectral distortion and the impact from 1H signals of the detergent in 13C-NMR spectroscopy experiments with HIGD1B, TMEM14A and TMEM141. Shigemi NMR tubes were used for solution NMR measurements. ‘Fingerprint’ spectra of the cell-free-expressed hIMPs, categorized as good, are shown in Figure 2 and Supplementary Figures 11 and 12. Cysteine mutagenesis and spin-labeling. For PRE experiments, 5–9 single-cysteine mutants were prepared for HIGD1A, HIGD1B, TMEM14A, FAM14B, TMEM141 and TMEM14C. Cysteine-free mutants were prepared for HIGD1B, TMEM14A and TMEM141 (Supplementary Table 3). The single cysteine mutants prepared for each of the six selected hIMPs are listed in Supplementary Table 3 and illustrated in Supplementary Figure 4. Positions for cysteine introduction were chosen based on the following criteria: mutations were located in regions containing helices, which were predicted by chemical shifts, and 3–4 residues adjoining the helix; each helix was labeled in at least two positions close to its ends; mutated residues had minimal structural and/or functional importance; the preferred amino acids for cysteine mutagenesis were serine, threonine and alanine; in the case there were no appropriate serine, threonine or alanine; residues close to the region of interest, valine, leucine, glutamine and aromatic tyrosine and phenylalanine residues were the second choice. Preservation of the structure in single-cysteine mutants was tested by TROSY-HSQC experiments as disruption of helical packing by the spin label would provide strong changes in chemical shifts and would diminish the overall quality of the TROSY spectra. All single-cysteine mutants gave minimal changes in TROSYHSQC spectra upon introduction of the mutations. Even though we followed the selection procedure described above, we still nature methods had few single-cysteine mutants for which satisfactory PRE data were not obtained. For example, PRE effect in 1-oxyl-(2,2,5,5tetramethyl-Δ3-pyrroline-3-methyl)methanethiosulfonate (MTSL)-labeled TMEM14A(S74C) affected only amide groups from neighboring residues (not more than 10 amino acids away from the labeled cysteine in the protein sequence). Other than that, PRE data from the paramagnetic label attached to a cysteine in non-transmembrane terminal helices require careful analysis, because of high mobility of the helix and attached label. The 15N-labeled single-cysteine mutants were prepared from 2–4 ml cell-free reaction mixture. To eliminate problems with (i) incomplete reduction of the spin-label in a detergent-solubilized protein and (ii) changes in chemical shifts caused by incorporation of a label, we used parallel labeling with structurally similar paramagnetic and diamagnetic labels as suggested in ref. 19. Every cysteine mutant was labeled with paramagnetic spin-label MTSL and with diamagnetic label 1-acetyl-(2,2,5,5-tetrametyl ∆3-pyrroline-3-methyl)methanethiosulfonate (DML) (both from Toronto Research Chemicals). The cysteine-free mutants or cysteine-free wild-type proteins were used as a control of nonspecific MTSL binding to the protein and/or detergent and were ‘labeled’ with MTSL only. For the labeling, the 15N-labeled NMR spectroscopy samples were split in half and supplemented with 5 mM MTSL or DML, solubilized in acetonitrile. After overnight incubation at room temperature, the excess of MTSL and DML was removed by washing in 5-kDa MWCO Vivaspin 2 concentrators (Sartorius Stedim Biotech). Labeled samples were washed 3 times by concentrating to 100 µl, resuspending in 2 ml NMR buffer and concentrating to 100 µl. After the third wash the samples were concentrated to 300 µl, supplemented with 5% D2O and 0.5 mM DSS and measured in a Shigemi NMR tube. NMR spectroscopy experiments. NMR spectra of hIMPs were recorded at 37 °C on a Bruker AVANCE 700 MHz spectrometer equipped with five radiofrequency channels and a triple-resonance cryoprobe with a shielded z-gradient coil. For the combinatorial assignment, [15N,1H]TROSY-HSQC and 15N,1H plane of the TROSY-HNCO12,37 were measured for each selectively 15N,13Clabeled sample. For the traditional assignment of backbone 1H, 15N and 13C resonances TROSY-based experiments, HNCA, HNCO38, HNCACB, HNCOCA, HNCOCACB and HNCACO39 as well as gradient-enhanced 3D 1H-15N-NOESY-TROSY (mixing time, 120 ms) were used. Partial side chain assignment was performed using 3D 1H-15N-NOESY-TROSY and 3D 1H-13CHSQC-NOESY–1H-15N-HSQC40 experiments. The PRE effect was measured using [15N,1H]TROSY-HSQC spectra collected for all cysteine mutants before spin labeling and after MTSL and DML labeling. Protein localization within LMPG micelles was checked by detection of a relaxation effect on [15N,1H]TROSYHSQC spectra of the hIMPs from water-soluble relaxation agent Gd3+-DOTA (Molecular Probes)41. HIGD1B was measured with different concentrations of Gd3+-DOTA (0 mM, 2.5 mM, 5.0 mM and 10 mM). After analysis of the relaxation effect using ratios of intensities in original (0 mM Gd3+-DOTA) and paramagnetic samples (Supplementary Fig. 3a), the 5.0 mM concentration was chosen to test the other hIMPs. The detected Gd3+-DOTA effect (Supplementary Fig. 3) confirmed the transmembrane topologies for the calculated hIMPs structures. Protein-detergent NOEs were derived using 3D 1H-15N-NOESY-TROSY experiments. doi:10.1038/nmeth.2033 npg © 2012 Nature America, Inc. All rights reserved. The spectra were collected with 200-ms mixing time for the 15N,2H-labeled proteins (to achieve full protonation of amides, the proteins were expressed and samples were prepared using H2O) in protonated LMPG. NMR signal assignment and spectra analysis. NMR spectra were transformed using Topspin (Bruker Biospin) and ProSA programs. Spectra analysis and assignment were performed using the CARA program. Combinatorial assignment using CDL strategy6 was used to accelerate assignment procedure. Built on the established principles of combinatorial assignment42,43, the CDL strategy generates a sequence-dependent labeling scheme for 5–8 samples6. The samples are expressed in the P-CF mode and selectively 15N,13C-labeled according to this scheme (see, for example, schemes for hIMPs in Supplementary Table 1). Analysis of the cross-peaks in paired TROSY-HSQC and 2D HNCO experiments allowed unambiguous assignment of those NH cross-peaks that correspond to unique amino acid pairs in the protein sequence. Other cross-peaks, with two or more possible assignments, are assigned to an amino acid type. The schemes for CDL assignment were calculated using MCCL program6 and consisted of 6, 6 and 5 samples for HIGD1A, HIGD1B and FAM14B, respectively (Supplementary Table 1). The selectivity of isotope labeling in cell-free reaction can be affected by amino acid scrambling. The most prominent examples are the pairs Gly-Ser, Asn-Gln, AspGlu and Asp-Asn. To avoid possible problems and facilitate CDL assignment, we modified the MCCL algorithm (http://sbl.salk. edu/combipro/) by incorporating user-defined identical labeling for any selected group of amino acids. In case of any uncertainty, the assignment can be verified using conventional sequential assignment methods. Calculation of PRE-based distance constraints. PRE distance constraints were introduced for distances between an amide proton and Cβ atom of residue, mutated to a cysteine for the paramagnetic labeling. Distance constraints were derived from the mea sured PRE effect using the procedures described in19,21,44,45. All the spectra were transformed in the same way and the intensities of 15N-1H cross-peaks in the MTSL (Ip) and DML (Id) samples were measured using the CARA program. The ratios of intensities (Ip/Id) were normalized against a set of 8–12 highest Ip/Id ratios, which were assumed to belong to cross-peaks unaffected by PRE. For TMEM141, the PRE distance constraints were derived by qualitative assessment of the Ip/Id ratios and were categorized based on intra- or inter-helical contacts between the spin label and the affected amide group as described45. PRE distance constraints for HIGD1A, HIGD1B, TMEM14A, FAM14B and TMEM14C were calculated using the modified SolomonBloembergen equation (equation (5) in ref. 21). The transverse relaxation rate enhancement was obtained from normalized intensity ratios (Ip/Id) as previously described, and the correlation time for the electron-nuclear spin interaction was estimated as the global rotational correlation time of the protein-detergent complex calculated using the Stokes-Einstein equation (Table 1). Such rough estimation of the correlation time is sufficient enough because, as mentioned earlier21, even moderate 20% error in estimation of the correlation time gives only ~0.5 Å error in calculated distance. For cross-peaks with the ratios below 0.15, no lower distance constraints were used, whereas upper constraints doi:10.1038/nmeth.2033 were set to 12 Å for distances between stable regions and to 15 Å if a spin label or amide group were located in flexible regions. For cross-peaks with the ratios above 0.9, only lower distance constraints equal to 25 Å were introduced. The upper distance constrains between flexible or unstructured regions were excluded from calculation. The upper and lower distance constraints for the peaks with Ip/Id ratios between 0.15 and 0.9 were generated from PRE-calculated distances using ± 4 Å margins. During structure calculation the margins were reduced to ± 3 Å for constraints between structured transmembrane regions if this change did not increase the Cyana penalty function. We assumed that the initial ± 4 Å margins in distance constraints are sufficient to cover the possible errors resulting from the use of a uniform correlation time, the uncertainty of the estimation of the intrinsic relaxation rates and the ‘r−6’-averaging of the nitroxide group motion. It was shown that the whole side chain of cysteine with attached MTSL (R1) has reduced mobility in both water-soluble and membrane proteins46,47. In the studied membrane leucine transporter46, the nitroxide rings interact with the hydrophobic protein surface, thus fixing the whole R1 side chain. We found that PRE-derived lower distance constraints are important for the calculation of helical bundle structures and especially for the determination of relative orientations of the transmembrane helices. As the only physical constraints of an α-helical bundle are van der Waals interactions, it has to have additional lower limit constrains from experimental data to prevent collapsing of the bundle. In contrast, others had reported that lower constraints were insignificant in the calculation of the β-barrel structure of OmpA19. This is not surprising because a β-barrel is restrained by the local geometry of inter-strand contacts and by the constant network of these contacts. As precision of the PRE-derived distance constraints is low, for successful structure calculation it is important to obtain as many meaningful constraints as possible. In theory, for every HN group from a given helical region the number of PRE-based distance constraints should be equal to the number of cysteine mutants used for spin-labeling. In reality, this number is lower due to signal overlapping and meaningless constraints like those between neighboring atoms within the same transmembrane helices. The average number of upper distance constraints per restrained residue ranged from 3.63 (HIGD1A) to 4.64 (FAM14B) for the studied hIMPs. According to ref. 48, approximate global fold can be determined with as few as 1.4 constraints per residue, whereas at least 3 constraints per residue are required for low- to medium-resolution NMR spectroscopy structures. Structure calculation and analysis. The 13Cα, 13Cβ and 13CO chemical shift deviations from random coil values were used to define backbone torsion angle restraints 49. Sequential distance constraints were derived from the integral intensities of NOE cross-peaks measured in 3D 15N-resolved TROSY-[1H;1H]NOESY (mixing time 120 ms). The hydrogen bond constraints were generated for the helical regions defined by chemical shift analysis. An interactive procedure, which included structure calculation by the CYANA program50 followed by the distance constraints refinement, was used to calculate the backbone spatial structures of the hIMPs. The structures were calculated using a simulated annealing protocol (1,000 high-temperature steps followed by 9,000–11,000 cooling down steps and 1,300 steps of a nature methods npg © 2012 Nature America, Inc. All rights reserved. conjugate gradient minimization) and the default CYANA force field. The summary of the constraints used in the calculation of the structures is presented in Table 1. Long-distance constraints used in structure calculation are shown in Figure 3b. The 20 conformers with the lowest target function of the last CYANA calculation cycle were selected from 200 calculated structures. The helical packing parameters, such as interhelical crossing angles and helical kinks, were derived for the final sets of 20 structures with the Helix Packing Pair51 and Molmol52 programs. The structures were visualized and analyzed in Molmol program; statistics for interatomic distances (average value and deviation) in the sets of structures were calculated using atomDistancer program. Cell culture. Human embryonic kidney cells HEK293T were grown in DMEM medium (Mediatech) supplemented with 10% FCS in humidified incubator at 37 °C with 5% CO2. Primary cultures of hippocampal neurons were prepared from 0–2-d-old Sprague Dawley rat pups using a modification of a previously described method53. Briefly, the hippocampi were dissected from brain and dissociated with papain (Worthington), and the neurons were plated at 25,000 cells/cm2 onto 12-mm glass cover slips (Warner Instruments) coated with 0.2 mg/ml poly-d-lysine (BD). Hippocampal neurons were cultured in Neurobasal medium supplement with B27, 100 U/ml streptomycin and 100 µg/ml penicillin (Invitrogen) at 37 °C and in 5% CO2 for 10–14 d. The medium was replaced the day after plating and twice weekly thereafter. All the procedures were approved by the Salk Institute’s Institutional Animal Care and Use Committee. Heterologous expression of hIMP proteins in human cell line. HEK293T cells were grown to 90–95% confluence and transiently transfected with DNA encoding for the indicated proteins using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s instructions. Forty-eight hours after transfection, cell culture medium was aspirated, and cells were washed twice with PBS. Cell were then placed on ice and lysed with mild lysis solution (Immunocatcher kit, CytoSignal Research Products) supplemented with protein inhibitors (Complete, Mini, Roche Applied Science). Immunofluorescence and imaging. Culture medium was aspirated from cell cultures, followed by a brief wash with PBS. Cells were then fixed with 4% paraformaldehyde solution in PBS for 10 min at room temperature, washed and permeabilized with 0.1% Triton-X in PBS (10 min). Unspecific binding was blocked by 30 min incubation with 3% BSA solution in PBS. Cells were then incubated with primary rabbit anti-HIGD1A antibody diluted 1:5,000 in blocking solution for 1 h, washed thoroughly with PBS and incubated over 1 h with secondary anti-rabbit Alexa Fluor 647–conjugated antibody (1:400 dilution, A-21244, Invitrogen). After the final wash with PBS, cells were mounted using ProLong Gold antifade reagent (Invitrogen) and imaged with a laser scanning confocal microscope (Zeiss LSM 710) using a 63× objective with oil immersion. The differential image contrast (DIC) was corrected using ImageJ Pseudoflatfield Filter followed by adjustment of brightness and contrast. Pictures of live HEK293T cell transfected with vectors encoding GFP fusion proteins were taken using a Nikon Eclipse TE300 inverted microscope equipped with a high-pressure mercury nature methods lamp and a GFP set of filters and beam splitters (Excitation filter HQ480/40x; dichroic mirror Q505LP, Emission filter HQ525/ 50m, Chroma) and a Nikon D70 digital camera. Modeling leverage calculations. The modeling leverage of the six hIMP NMR spectroscopy structures was estimated by the ModPipe, comparative modeling pipeline54 accessible through the ModWeb web server (http://salilab.org/modweb/) (Supplementary Table 5). We relied on the ModWeb option that accepts a protein structure as input, calculates a multiple sequence profile and identifies all homologous sequences in the UniProtKB database55, followed by modeling these homologs based on the user-provided structure. These models are available in ModBase through a summary page (http://modbase.compbio.ucsf.edu/ modbase-cgi/model_leverage.cgi?type=master_salk). On average, each structure allowed us to model with relatively high accuracy 171 related unique protein sequences, based on more than 30% sequence identity and using at least 50% of the residues in the structures as templates. Another 114 protein sequences on average could also be modeled, but at lower accuracy, primarily because of the target-template alignment errors. 32. Zheng, L., Baumann, U. & Reymond, J.L. An efficient one-step sitedirected and site-saturation mutagenesis protocol. Nucleic Acids Res. 32, e115 (2004). 33. Shi, C. et al. Purification and characterization of a recombinant G-proteincoupled receptor, Saccharomyces cerevisiae Ste2p, transiently expressed in HEK293 EBNA1 cells. Biochemistry 44, 15705–15714 (2005). 34. Klammt, C., Schwarz, D., Dotsch, V. & Bernhard, F. Cell-free production of integral membrane proteins on a preparative scale. Methods Mol. Biol. 375, 57–78 (2007). 35. Ichetovkin, I.E., Abramochkin, G. & Shrader, T.E. Substrate recognition by the leucyl/phenylalanyl-tRNA-protein transferase. Conservation within the enzyme family and localization to the trypsin-resistant domain. J. Biol. Chem. 272, 33009–33014 (1997). 36. Savage, D.F. et al. Cell-free complements in vivo expression of the E. coli membrane proteome. Protein Sci. 16, 966–976 (2007). 37. Riek, R. et al. Solution NMR techniques for large molecular and supramolecular structures. J. Am. Chem. Soc. 124, 12144–12153 (2002). 38. Salzmann, M. et al. TROSY in triple-resonance experiments: new perspectives for sequential NMR assignment of large proteins. Proc. Natl. Acad. Sci. USA 95, 13585–13590 (1998). 39. Salzmann, M. et al. TROSY-type triple-resonance experiments for sequential NMR assignments of large proteins. J. Am. Chem. Soc. 121, 844–848 (1999). 40. Diercks, T., Coles, M. & Kessler, H. An efficient strategy for assignment of cross-peaks in 3D heteronuclear NOESY experiments. J. Biomol. NMR 15, 177–180 (1999). 41. Hilty, C., Wider, G., Fernandez, C. & Wuthrich, K. Membrane protein-lipid interactions in mixed micelles studied by NMR spectroscopy with the use of paramagnetic reagents. ChemBioChem 5, 467–473 (2004). 42. Yabuki, T. et al. Dual amino acid-selective and site-directed stable-isotope labeling of the human c-Ha-Ras protein by cell-free synthesis. J. Biomol. NMR 11, 295–306 (1998). 43. Kainosho, M. & Tsuji, T. Assignment of the three methionyl carbonyl carbon resonances in Streptomyces subtilisin inhibitor by a carbon-13 and nitrogen-15 double-labeling technique. A new strategy for structural studies of proteins in solution. Biochemistry 21, 6273–6279 (1982). 44. Van Horn, W.D., Beel, A.J., Kang, C. & Sanders, C.R. The impact of window functions on NMR-based paramagnetic relaxation enhancement measurements in membrane proteins. Biochim. Biophys. Acta 1798, 140–149 (2010). 45. Roosild, T.P. et al. NMR structure of Mistic, a membrane-integrating protein for membrane protein expression. Science 307, 1317–1321 (2005). 46. Kroncke, B.M., Horanyi, P.S. & Columbus, L. Structural origins of nitroxide side chain dynamics on membrane protein alpha-helical sites. Biochemistry 49, 10045–10060 (2010). 47. Langen, R., Oh, K.J., Cascio, D. & Hubbell, W.L. Crystal structures of spin labeled T4 lysozyme mutants: implications for the interpretation of EPR spectra in terms of structure. Biochemistry 39, 8396–8405 (2000). doi:10.1038/nmeth.2033 52. Koradi, R., Billeter, M. & Wuthrich, K. MOLMOL: a program for display and analysis of macromolecular structures. J. Mol. Graph. 14, 51–55 (1996). 53. Brewer, G.J., Torricelli, J.R., Evege, E.K. & Price, P.J. Optimized survival of hippocampal neurons in B27-supplemented Neurobasal, a new serum-free medium combination. J. Neurosci. Res. 35, 567–576 (1993). 54. Pieper, U. et al. ModBase, a database of annotated comparative protein structure models, and associated resources. Nucleic Acids Res. 39, D465–D474 (2011). 55. Bairoch, A. et al. The Universal Protein Resource (UniProt). Nucleic Acids Res. 33, D154–D159 (2005). npg © 2012 Nature America, Inc. All rights reserved. 48. Clore, G.M., Robien, M.A. & Gronenborn, A.M. Exploring the limits of precision and accuracy of protein structures determined by nuclear magnetic resonance spectroscopy. J. Mol. Biol. 231, 82–102 (1993). 49. Lubingbuhl, P., Szyperski, T. & Wuthrich, K. Statistical basis for the use of 13Ca chemical shifts in protein structure determination. J. Magn. Reson. B. 109, 229–233 (1995). 50. Guntert, P. Automated NMR structure calculation with CYANA. Methods Mol. Biol. 278, 353–378 (2004). 51. Dalton, J.A., Michalopoulos, I. & Westhead, D.R. Calculation of helix packing angles in protein structures. Bioinformatics 19, 1298–1299 (2003). doi:10.1038/nmeth.2033 nature methods
